Anthropic’s long-awaited S-1, filed Monday ahead of its planned Nasdaq debut under the ticker ANT, shows how quickly customers are spending on its newest models.
According to the filing, models released in the six months leading up to September 30 accounted for more than 80% of the company’s third-quarter revenue. Anthropic released eight widely available Claude models between April and September 2026, plus three restricted-access Mythos models. The filing doesn’t say how much revenue each generated; models launched in late September had only days to contribute to the quarter.
But that raises an interesting question for developers: Are newer models delivering better results for developers’ tasks, and what do those results cost?
Are newer models delivering better results for developers’ tasks, and what do those results cost?
How new models compound
The filing suggests there’s a reason newer models account for so much of Anthropic’s revenue. The company says customer spending is closely tied to how often it releases new models and what those models can do. Each improvement naturally gives customers more reasons to use Claude, while more capable models attract new customers and encourage existing ones to use Claude more and spend more in the process.
That spending helps fund the next round of development. Anthropic says in the filing that the company relies on growing revenue to pay for the research, computing power, and technical talent needed to build more capable models. Those improvements can then attract more customers and drive additional spending, creating a cycle in which each generation of Claude helps fund the next.
Improvements can then attract more customers and drive additional spending, creating a cycle in which each generation of Claude helps fund the next.
Did demand drive releases?
The filing shows how important newer models are to Anthropic’s revenue, but it doesn’t establish whether customer demand has influenced the company’s faster release schedule in 2026. While the two may be connected, Anthropic doesn’t say that increased spending is what drove it to release models more frequently.
Revenue, not customer migration
The 80% figure also doesn’t tell us how many customers switched to newer models. A relatively small group of customers spending heavily on the latest releases could account for much of that revenue.
The filing offers some context for how much spending can vary. Anthropic’s heaviest Claude Code enterprise users spend roughly 20 times more than other enterprise users. That doesn’t tell us which models they’re using, but it shows why revenue alone can’t tell us how many customers have made the switch.
Choosing the newest model
Developers building coding agents, retrieval pipelines and automation tools may default to the most powerful model without first checking whether a smaller or cheaper alternative could handle the job. That spending adds up. Reuters reported — citing attendees at Anthropic’s October 14 investor meeting — that the company generated $16.8 billion in third-quarter revenue.
Anthropic reportedly generated $16.8 billion in third-quarter revenue.
Whether developers actually need the latest models or simply prefer them, Anthropic benefits from the spending. With models released since April accounting for more than 80% of its third-quarter revenue, Anthropic has a strong financial incentive to keep improving Claude.
Anthropic has also called for pacing frontier AI development so safety work can keep up with growing model capabilities. The filing does not show whether commercial demand has affected that approach.
The post Most of Anthropic’s revenue comes from models that barely existed six months ago appeared first on The New Stack.